This paper describes a coupled fire-atmosphere model that uses a sophisticated high-resolution non-hydrostatic numerical mesoscale model to predict the local winds which are then used as input to the prediction of fire spread. The heat and moisture fluxes from the fire are then fed back to the dynamics, allowing the fire to influence its own mesoscale winds that in turn affect the fire behavior. This model is viewed as a research model and as such requires a fireline propagation scheme that systematically converges with increasing spatial and temporal resolution. To achieve this, a local contour advection scheme was developed to track the fireline using four tracer particles per fuel cell, which define the area of burning fuel. Using the dynamically predicted winds along with the terrain slope and fuel characteristics, algorithms from the BEHAVE system are used to predict the spread rates. A mass loss rate calculation, based on results of the BURNUP fuel burnout model, is used to treat heat exchange between the fire and atmosphere. Tests were conducted with the uncoupled model to test the fire-spread algorithm under specified wind conditions for both spot and line fires. Using tall grass and chaparral, line fires were simulated employing the full fire-atmosphere coupling. Results from two of these experiments show the effects of fire propagation over a small hill. As with previous coupled experiments, the present results show a number of features common to real fires. For example, we show how the well-recognized elliptical fireline shape is a direct result of fire-atmosphere interactions that produce the 'heading', 'flanking', and 'backing' regions of a wind-driven fire with their expected behavior. And, we see how perturbations upon this shape sometimes amplify to become fire whirls along the flanks, which are transported to the head of the fire where they may interact to produce erratic fire behavior.